🚀 Riksdagsmonitor — Future State Diagrams

🔮 Advanced System Behavior, Lifecycles and State Transitions
🎯 Static-First Enrichment → AWS Serverless · AI Content Lifecycles · Predictive & Streaming States

📋 Document Owner: CEO | 📄 Version: 3.0 | 📅 Last Updated: 2026-05-31 (UTC)
🔄 Review Cycle: Quarterly | ⏰ Next Review: 2026-08-31
🏢 Owner: Hack23 AB (Org.nr 559534-7807) | 🏷️ Classification: Public

🎯 Purpose

This document projects the state-transition and lifecycle evolution of Riksdagsmonitor across a ten-year horizon (2026–2037), building directly on the freshly-refreshed current-state models in STATEDIAGRAM.md. It is organised around three horizons so that every future state machine is anchored to a real, shipped baseline rather than speculation:

1. 🟢 v1.x baseline (today, shipped) — the article lifecycle (analysis-gate → draft → translate → render → publish), data-sync states, and dashboard build states that already run in production on the static AWS CloudFront + multi-region S3 estate.
2. 🔵 v2.0 (2026–2027) — static, but deeper — enriched build/publish lifecycles for party-focused dashboards and OSINT/INTOP intelligence products, including evidence-grading and anomaly-detection states, all still emitted as static artifacts. No runtime migration.
3. 🟣 v3.0+ (2028–2037) — AWS serverless — AI content-generation lifecycles on Amazon Bedrock, the predictive-model lifecycle (train → validate → deploy → monitor → retrain), real-time streaming states (Kinesis), election-forecast model states, multi-parliament coordination, Bedrock Knowledge Base ingestion/refresh states, Cognito session/auth states and the public API request lifecycle.

A standing AI/LLM Model Evolution Lifecycle (2026–2037) state machine translates the Hack23 AI Model Evolution roadmap into state terms and threads it through all three horizons.

Lens of this document: system state transitions & lifecycles. For the conceptual map see FUTURE_MINDMAP.md; for processes FUTURE_FLOWCHART.md; for structure FUTURE_ARCHITECTURE.md; for data FUTURE_DATA_MODEL.md; for strategy FUTURE_SWOT.md.

📚 Architecture Documentation Map

📑 Table of Contents

• Executive Summary
• Horizon 1 — v1.x baseline (shipped)
  • 1. Article Lifecycle (analysis-gate → publish)
  • 2. Data-Sync Lifecycle
  • 3. Dashboard Build Lifecycle
• Horizon 2 — v2.0 static enrichment (2026–2027)
  • 4. Party-Dashboard Build & Publish Lifecycle
  • 5. Evidence-Grading State Machine
  • 6. Anomaly-Detection State Machine
  • 7. OSINT / INTOP Scorecard Lifecycle
• Horizon 3 — v3.0+ AWS serverless (2028–2037)
  • 8. Bedrock AI Content-Generation Lifecycle
  • 9. Predictive-Model Lifecycle (train → retrain)
  • 10. Real-Time Streaming States (Kinesis)
  • 11. Election-Forecast Model States
  • 12. Multi-Parliament Coordination States
  • 13. Bedrock Knowledge Base Ingestion / Refresh States
  • 14. Cognito Session / Auth States
  • 15. API Request Lifecycle (API Gateway → Lambda)
• Cross-Horizon
  • 16. AI/LLM Model Evolution Lifecycle (2026–2037)
  • 17. IMF Cache → Aurora Lifecycle Evolution
  • 18. 🛰️ Political-Intelligence Capability State Machines (to 2037)
• Future State Summary
• ISMS Control Mapping
• Document Control
• Hack23 Ecosystem

📋 Executive Summary

Riksdagsmonitor's behavioural future is a deliberate two-step: first deepen the static platform (v2.0) without changing its operational substrate, then migrate the runtime to AWS serverless (v3.0+) once the intelligence products justify managed AI and live APIs. This document models that journey purely in state-machine terms so engineers, auditors and stakeholders can reason about what states an entity can occupy, what guards each transition, and what invariants hold at every horizon.

Invariants that hold across all horizons: public-data-only collection; political opinions processed under GDPR Art. 9(2)(e) (manifestly made public) and 9(2)(g) (substantial public interest); strict party neutrality; no surveillance/psyops framing; immutable, audit-preserved provenance; human-in-the-loop governance per the Hack23 AI Policy.

🟢 Horizon 1 — v1.x Baseline States (Shipped)

These machines reflect what runs today. They are reproduced here (refreshed, not invented) so each future horizon visibly extends a real baseline. The authoritative current models live in STATEDIAGRAM.md.

1. 📰 Horizon 1: Article Lifecycle (analysis-gate → publish)

The autonomous newsroom (14 gh-aw agentic workflows, Claude Opus 4.8 authoring / Sonnet 4.6 translation, zero human editors) drives every article through a blocking analysis gate before any prose exists.

stateDiagram-v2
    [*] --> ArtifactsEmpty

    ArtifactsEmpty --> Populating: Agentic workflow writes 9–23 analysis artifacts
    Populating --> AnalysisGate: All required artifacts present
    Populating --> Incomplete: Timeout or agent error

    AnalysisGate --> Draft: analysis-gate.ts checks 1–9b PASS
    AnalysisGate --> Populating: Gate FAIL (retry artifacts)

    Draft --> Translate: Swedish content + data-translate markers
    Draft --> GenError: Generation failed

    Translate --> Validate: 14 languages emitted
    Translate --> TranslateFailed: Hallucination / context loss

    Validate --> Render: No residual data-translate markers
    Validate --> ValidateFailed: Markers detected

    Render --> Publish: HTML built · RSS · sitemap · hreflang
    Publish --> Live: PR merged · CloudFront invalidated
    Live --> [*]

    Incomplete --> [*]: Workflow terminates with error
    GenError --> Draft: Fix and retry
    TranslateFailed --> Draft: Retry translation
    ValidateFailed --> Translate: Fix translations
    Live --> Rollback: Production issue
    Rollback --> Draft: Recreate article

    note right of AnalysisGate
        Single blocking gate (05-analysis-gate.md)
        Evidence, Mermaid colour, Pass-2 proof,
        political classification, agency evidence
    end note

    note right of Live
        Immutable · SHA-256 hash recorded
        Cannot transition back to Draft
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Draft,Live,Publish ok
    class AnalysisGate,Validate,Translate warn
    class GenError,TranslateFailed,ValidateFailed,Incomplete,Rollback err
Copy

2. 🔄 Horizon 1: Data-Sync Lifecycle

Open-data ingestion from the Riksdagen API, Regeringen (via g0v.se), SCB, IMF, World Bank and the CIA platform (Hack23) JSON/CSV exports — 349 current MPs, 2,494 historical politicians (1971–2024), 3.5M+ votes, 109,000+ documents.

stateDiagram-v2
    [*] --> Fresh

    Fresh --> SlightlyStale: Age 1–3 days
    SlightlyStale --> Fresh: Refresh OK
    SlightlyStale --> Stale: Age > 3 days

    Stale --> Refreshing: Scheduled sync fires
    Stale --> Critical: Age > 7 days
    Critical --> Refreshing: Emergency sync

    Refreshing --> Fetching: MCP / export pull starts
    Fetching --> SchemaValidate: Payload received
    Fetching --> RefreshFailed: Network / 5xx

    SchemaValidate --> Fresh: Validation passes · SHA-256 pinned
    SchemaValidate --> RefreshFailed: Validation fails

    RefreshFailed --> CacheFallback: Serve last good vintage
    CacheFallback --> Stale: Stale banner shown
    CacheFallback --> Expired: No valid cache

    Expired --> OperatorAlert: Data > 14 days
    OperatorAlert --> Refreshing: Manual retry

    note right of Fresh
        Sources: riksdag-regering · scb · imf ·
        world-bank · CIA exports
        Banner: "Data current"
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Fresh ok
    class SlightlyStale,Stale,Refreshing,Fetching,CacheFallback warn
    class Critical,RefreshFailed,Expired,OperatorAlert err
Copy

3. 📊 Horizon 1: Dashboard Build Lifecycle

~11 functional dashboards (party, ministry, anomaly-detection, seasonal-patterns, pre-election, politician, …) are lazy-loaded TypeScript modules (Chart.js / D3.js) compiled by Vite and shipped as static assets — no JS frameworks for content.

stateDiagram-v2
    [*] --> SourceData

    SourceData --> Transform: data-sync emits Fresh datasets
    Transform --> Compile: JSON shaped for dashboards
    Compile --> BundleSplit: Vite builds TS modules
    BundleSplit --> AssetEmit: Lazy chunks per dashboard

    AssetEmit --> SmokeCheck: Static assets emitted
    SmokeCheck --> Deployable: Chart.js / D3 render verified
    SmokeCheck --> BuildFailed: Render or type error

    Deployable --> CDNPublish: Pushed to S3
    CDNPublish --> Cached: CloudFront warm
    Cached --> [*]

    BuildFailed --> Transform: Fix data / module

    note right of BundleSplit
        Lazy-loaded modules · WCAG 2.1 AA
        14 languages · cyberpunk theme
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Deployable,Cached,CDNPublish ok
    class Transform,Compile,BundleSplit,SmokeCheck warn
    class BuildFailed err
Copy

🔵 Horizon 2 — v2.0 Static Enrichment (2026–2027)

Strategic guardrail: the substrate does not change. Every state machine below still terminates in a static artifact published to CloudFront + S3. What deepens is the intelligence quality of the build pipeline — party analytics, graded evidence and anomaly detection — driven by the agentic newsroom.

4. 🏛️ Horizon 2: Party-Dashboard Build & Publish Lifecycle

Better dashboards covering the political landscape with a focus on parties: party cohesion, coalition dynamics, bloc alignment, party-vs-party comparison and agenda tracking.

stateDiagram-v2
    [*] --> PartyDataReady

    PartyDataReady --> CohesionCompute: Votes + memberships loaded
    CohesionCompute --> BlocAlignment: Party-discipline scored
    BlocAlignment --> CoalitionMatrix: Bloc vectors built
    CoalitionMatrix --> AgendaTrack: Coalition-probability grid ready

    AgendaTrack --> PartyAssemble: Agenda shifts tagged
    PartyAssemble --> A11yGate: Comparative panels composed

    A11yGate --> I18nGate: WCAG 2.1 AA passes
    A11yGate --> PartyAssemble: Accessibility defect

    I18nGate --> StaticEmit: 14 languages + RTL valid
    I18nGate --> PartyAssemble: Missing translation

    StaticEmit --> CDNPublish: Lazy TS modules built
    CDNPublish --> LivePartyDashboard: CloudFront warm
    LivePartyDashboard --> RefreshCycle: Next data-sync
    RefreshCycle --> CohesionCompute: Recompute on Fresh data
    LivePartyDashboard --> [*]

    note right of CohesionCompute
        Rice cohesion index · whip-defiance rate
        Evidence: dok_id + vote counts (no opinion mining)
    end note

    note right of CoalitionMatrix
        Power × interest × position
        Neutral, all 8 parties symmetric
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class LivePartyDashboard,CDNPublish,StaticEmit ok
    class CohesionCompute,BlocAlignment,CoalitionMatrix,AgendaTrack,PartyAssemble,A11yGate,I18nGate warn
Copy

5. 🏅 Horizon 2: Evidence-Grading State Machine

Every claim emitted by an OSINT product carries a source grade. This lifecycle promotes or demotes evidence before it can appear in a published artifact — the state-machine encoding of the 05-analysis-gate.md evidence checks.

stateDiagram-v2
    [*] --> Unverified

    Unverified --> SourceCheck: Claim asserts dok_id / actor / vote
    SourceCheck --> PrimaryConfirmed: Primary source resolves (Riksdagen / Regeringen / SCB / IMF)
    SourceCheck --> SecondaryOnly: Only secondary corroboration
    SourceCheck --> Rejected: No resolvable source

    PrimaryConfirmed --> GradeA: Single authoritative primary
    PrimaryConfirmed --> GradeAPlus: ≥2 independent primaries agree
    SecondaryOnly --> GradeB: Corroborated, not primary
    SecondaryOnly --> GradeC: Single secondary / uncertain

    GradeAPlus --> Publishable: Full confidence
    GradeA --> Publishable: Full confidence
    GradeB --> PublishableFlagged: Confidence downgraded note
    GradeC --> PublishableFlagged: Explicit uncertainty label
    Rejected --> Quarantine: Excluded from article

    Publishable --> [*]
    PublishableFlagged --> [*]
    Quarantine --> Unverified: Re-evaluate on new data

    note right of GradeAPlus
        ACH-compatible: competing hypotheses
        each carry their own evidence grade
    end note

    note right of Quarantine
        No claim ships without a grade.
        Generic statements are rejected.
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class GradeAPlus,GradeA,Publishable ok
    class SecondaryOnly,GradeB,GradeC,PublishableFlagged warn
    class Rejected,Quarantine err
Copy

6. 🔎 Horizon 2: Anomaly-Detection State Machine

The anomaly-detection dashboard flags statistically unusual voting, attendance or spending patterns — surfaced as signals, never accusations, and always traced to public records.

stateDiagram-v2
    [*] --> Baselining

    Baselining --> Monitoring: Historical distributions modelled
    Monitoring --> SignalCandidate: Metric breaches z-score threshold
    Monitoring --> Monitoring: Within expected band

    SignalCandidate --> ContextEnrich: Outlier captured
    ContextEnrich --> FalsePositive: Explained by known event
    ContextEnrich --> Corroborating: Unexplained deviation

    FalsePositive --> Monitoring: Tune threshold

    Corroborating --> EvidenceGrade: Cross-check primary sources
    EvidenceGrade --> SignalConfirmed: Grade A/A+ evidence
    EvidenceGrade --> SignalWeak: Grade B/C evidence

    SignalConfirmed --> Published: Neutral signal card emitted
    SignalWeak --> Watchlist: Held, not published
    Watchlist --> Corroborating: New data arrives

    Published --> [*]

    note right of SignalCandidate
        Examples: vote defection spike,
        attendance collapse, budget-line jump
    end note

    note right of Published
        Framed as "signal for inquiry"
        Never asserts intent or wrongdoing
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class SignalConfirmed,Published ok
    class Baselining,Monitoring,SignalCandidate,ContextEnrich,Corroborating,EvidenceGrade,SignalWeak,Watchlist warn
    class FalsePositive err
Copy

7. 🕵️ Horizon 2: OSINT / INTOP Scorecard Lifecycle

INTOP scorecards (attendance, voting discipline, productivity, committee contribution) and network/temporal/geospatial OSINT products, advancing political-intelligence quality while staying static.

stateDiagram-v2
    [*] --> Collection

    Collection --> NetworkBuild: Public records gathered (MCP)
    NetworkBuild --> TemporalModel: Co-sponsorship / vote graph built
    TemporalModel --> ScoreCompute: Time-series patterns extracted
    ScoreCompute --> EvidenceGradeRef: Scorecard metrics derived

    EvidenceGradeRef --> PeerReviewSim: Each metric graded (§5)
    PeerReviewSim --> ScorecardDraft: Red-team / devil's-advocate pass
    PeerReviewSim --> ScoreCompute: Bias or gap detected

    ScorecardDraft --> StaticEmit: Neutrality + methodology disclosed
    StaticEmit --> LiveScorecard: Published in 14 languages
    LiveScorecard --> Refresh: Scheduled re-score
    Refresh --> Collection
    LiveScorecard --> [*]

    note right of NetworkBuild
        Centrality · clustering · bridging
        Public co-sponsorship & voting only
    end note

    note right of PeerReviewSim
        Structured tradecraft: ACH, SWOT,
        PESTLE, Red Team applied per product
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    class LiveScorecard,StaticEmit ok
    class Collection,NetworkBuild,TemporalModel,ScoreCompute,EvidenceGradeRef,PeerReviewSim,ScorecardDraft,Refresh warn
Copy

🟣 Horizon 3 — v3.0+ AWS Serverless (2028–2037)

The strategic migration: all-in AWS serverless, zero infrastructure to manage — no Kubernetes, no containers — managed AI + serverless only, AWS Well-Architected, multi-region. The state machines below model the new runtime behaviours this substrate unlocks.

8. 🤖 Horizon 3: Bedrock AI Content-Generation Lifecycle

Article generation moves from build-time gh-aw workflows into Amazon Bedrock foundation models and Bedrock Agents, orchestrated by Step Functions and grounded by Bedrock Knowledge Bases (RAG over the political corpus).

stateDiagram-v2
    [*] --> EventDetected

    EventDetected --> ContextRetrieval: EventBridge rule fires
    ContextRetrieval --> Generation: Knowledge Base RAG grounded
    ContextRetrieval --> InsufficientContext: KB recall too low

    InsufficientContext --> ContextRetrieval: Widen retrieval window

    Generation --> GroundingCheck: Bedrock model drafts article
    Generation --> ModelError: Invocation failed

    GroundingCheck --> QualityEval: Citations resolve to corpus
    GroundingCheck --> Generation: Hallucination detected (retry)

    QualityEval --> MultiModal: Self-eval score ≥ 0.8
    QualityEval --> Generation: Score < 0.8 (revise)

    MultiModal --> Translate14: Text body finalised
    Translate14 --> GovernanceGate: 14 languages + RTL produced

    GovernanceGate --> Publish: AI Policy checks pass (neutrality, provenance)
    GovernanceGate --> HumanEscalation: Sensitive-topic flag

    HumanEscalation --> Publish: Reviewer clears
    HumanEscalation --> Generation: Reviewer rejects

    Publish --> Indexed: Stored DynamoDB + OpenSearch
    Indexed --> Archived: Age > 90 days
    Archived --> [*]

    ModelError --> FallbackTemplate: Max retries exceeded
    FallbackTemplate --> GovernanceGate: Deterministic template content

    note right of ContextRetrieval
        Bedrock Knowledge Base (RAG)
        OpenSearch Serverless vectors
        Provenance preserved per chunk
    end note

    note right of GovernanceGate
        Human-in-the-loop only for
        flagged sensitive topics — else
        autonomous within policy guardrails
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Publish,Indexed ok
    class ContextRetrieval,Generation,GroundingCheck,QualityEval,MultiModal,Translate14,GovernanceGate,HumanEscalation warn
    class ModelError,InsufficientContext,FallbackTemplate err
Copy

9. 📈 Horizon 3: Predictive-Model Lifecycle (train → retrain)

Forecasting models (vote prediction, coalition probability) run a disciplined MLOps lifecycle on serverless training and Bedrock/SageMaker-class inference, with shadow and canary gates before any public exposure.

stateDiagram-v2
    [*] --> DataIngestion

    DataIngestion --> FeatureEngineering: Raw votes / polls / history (50+ yrs)
    FeatureEngineering --> Training: Features materialised (Timestream + Aurora)
    Training --> Validation: Training run complete

    Validation --> ShadowMode: Holdout R² / accuracy ≥ threshold
    Validation --> Training: Below threshold (tune)

    ShadowMode --> CanaryRollout: 7-day shadow stable
    ShadowMode --> Rollback: Shadow drift

    state CanaryRollout {
        [*] --> Traffic5
        Traffic5 --> Traffic25: Metrics green
        Traffic25 --> Traffic50: Metrics green
        Traffic50 --> [*]: All green
    }

    CanaryRollout --> FullDeployment: Canary cleared
    CanaryRollout --> Rollback: Metric regression

    FullDeployment --> Monitoring: Model serving via API Gateway
    Monitoring --> Retraining: Drift / decay detected
    Monitoring --> FullDeployment: Stable
    Retraining --> DataIngestion: New training cycle

    Rollback --> FullDeployment: Revert to last good model

    note right of Validation
        Statistical significance p < 0.05
        Backtested against historical cycles
        Fairness / neutrality checks
    end note

    note right of Monitoring
        CloudWatch + EventBridge drift alarms
        Confidence intervals always published
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class FullDeployment,Monitoring ok
    class DataIngestion,FeatureEngineering,Training,Validation,ShadowMode,CanaryRollout,Retraining warn
    class Rollback err
Copy

10. 🌊 Horizon 3: Real-Time Streaming States (Kinesis)

Live parliamentary sessions stream through Amazon Kinesis Data Streams, enriched by Lambda and broadcast to API consumers — replacing the build-time batch model with sub-second push.

stateDiagram-v2
    [*] --> Idle

    Idle --> Provisioning: Parliamentary session scheduled
    Provisioning --> Streaming: Kinesis shards open

    Streaming --> Processing: Record arrives
    Processing --> Enrichment: Lambda parses event
    Enrichment --> Broadcasting: Context + grade attached
    Broadcasting --> Streaming: Pushed to subscribers

    Streaming --> Buffering: Throughput burst
    Buffering --> Reshard: Shard limit hit
    Reshard --> Streaming: Capacity scaled
    Buffering --> Processing: Buffer flushed

    Processing --> DeadLetter: Processing failure
    DeadLetter --> Processing: Replay from checkpoint

    Streaming --> Degraded: Consumer lag critical
    Degraded --> PollingFallback: Serve cached REST snapshot
    PollingFallback --> Streaming: Lag recovered

    Streaming --> SessionEnd: Session adjourns
    SessionEnd --> Archiving: Flush to Timestream + S3
    Archiving --> Idle: Archive sealed

    note right of Streaming
        Kinesis Data Streams · Lambda consumers
        Sub-second target · auto-resharding
    end note

    note right of Degraded
        Graceful degradation to cached REST
        Subscribers notified of mode change
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Streaming,Broadcasting ok
    class Idle,Provisioning,Processing,Enrichment,Buffering,Reshard,SessionEnd,Archiving warn
    class DeadLetter,Degraded,PollingFallback err
Copy

11. 🗳️ Horizon 3: Election-Forecast Model States

Election forecasting runs a cycle-aware state machine with an ethical blackout on election day — predictions are never published while polls are open.

stateDiagram-v2
    [*] --> PreElection

    PreElection --> Calibration: Cycle begins (T-1460d window)
    Calibration --> WeeklyForecast: > 30 days to election
    Calibration --> DailyForecast: ≤ 30 days to election

    WeeklyForecast --> Calibration: New week of data
    DailyForecast --> Calibration: New day of data

    DailyForecast --> ElectionDayBlackout: Polls open
    ElectionDayBlackout --> LiveTracking: Polls close
    LiveTracking --> ResultComparison: Official results stream in
    ResultComparison --> PostElection: Final results certified

    PostElection --> AccuracyAnalysis: Forecast vs outcome scored
    AccuracyAnalysis --> ModelRetraining: Brier / calibration lessons
    ModelRetraining --> ArchivedCycle: Tuned for next cycle
    ArchivedCycle --> [*]

    note right of WeeklyForecast
        Monte Carlo: 10,000+ simulations
        90% / 95% confidence intervals
        Coalition-probability matrix (12 leaves)
    end note

    note right of ElectionDayBlackout
        No forecasts published while voting open.
        Ethical commitment — avoid voter influence.
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class WeeklyForecast,DailyForecast,PostElection ok
    class PreElection,Calibration,LiveTracking,ResultComparison,AccuracyAnalysis,ModelRetraining,ArchivedCycle warn
    class ElectionDayBlackout err
Copy

12. 🌍 Horizon 3: Multi-Parliament Coordination States

Federated expansion to Nordic and EU parliaments (Folketing, Storting, Eduskunta, European Parliament) coordinated via Step Functions, reusing the European Parliament MCP.

stateDiagram-v2
    [*] --> Initialization

    state Initialization {
        [*] --> Riksdag
        [*] --> Folketing
        [*] --> Storting
        [*] --> Eduskunta
        [*] --> EUParliament
    }

    Initialization --> DataSync: All sources configured
    DataSync --> SchemaNormalization: Raw data collected
    SchemaNormalization --> EntityResolution: Unified schema
    EntityResolution --> GraphMerge: Cross-parliament entities matched
    GraphMerge --> CrossCountryAnalysis: Neptune graph updated

    CrossCountryAnalysis --> ComparativeProduct: Analysis complete
    ComparativeProduct --> Monitoring: Product live (static or API)
    Monitoring --> DataSync: Refresh cycle

    DataSync --> SourceError: A parliament API unavailable
    SourceError --> PartialServe: Serve available parliaments
    PartialServe --> DataSync: Source restored

    note right of SchemaNormalization
        Heterogeneous formats: XML / JSON / HTML / PDF
        Language normalisation across systems
    end note

    note right of CrossCountryAnalysis
        Comparative cohesion, productivity,
        diversity, budget priorities — neutral
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class ComparativeProduct,Monitoring ok
    class DataSync,SchemaNormalization,EntityResolution,GraphMerge,CrossCountryAnalysis warn
    class SourceError,PartialServe err
Copy

13. 📚 Horizon 3: Bedrock Knowledge Base Ingestion / Refresh States

The Bedrock Knowledge Base is the RAG memory of the platform. Its corpus (109,000+ documents and growing) must ingest, embed, version and refresh while preserving provenance for every chunk.

stateDiagram-v2
    [*] --> Empty

    Empty --> Ingesting: New / updated source documents arrive
    Ingesting --> Chunking: Documents fetched (S3 data source)
    Chunking --> Embedding: Provenance-tagged chunks created
    Embedding --> Indexing: Vectors generated (Titan / model)
    Indexing --> Ready: Written to OpenSearch Serverless

    Ready --> Serving: RAG queries answered
    Serving --> Ready: Query complete

    Ready --> RefreshDue: Source vintage changed / schedule
    RefreshDue --> IncrementalSync: Delta detected
    IncrementalSync --> Embedding: Re-embed changed chunks
    RefreshDue --> Ready: No changes

    Embedding --> EmbedError: Model / quota failure
    EmbedError --> Backoff: Exponential retry
    Backoff --> Embedding: Retry
    EmbedError --> StaleServe: Serve prior index (flagged)
    StaleServe --> Ready: Refresh recovers

    note right of Chunking
        Each chunk keeps dok_id / URL / vintage
        so RAG answers remain citable
    end note

    note right of StaleServe
        KB never silently serves stale corpus —
        downgraded-confidence annotation applied
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Ready,Serving,Indexing ok
    class Empty,Ingesting,Chunking,Embedding,RefreshDue,IncrementalSync warn
    class EmbedError,Backoff,StaleServe err
Copy

14. 🔐 Horizon 3: Cognito Session / Auth States

Amazon Cognito provides identity for personalization and API consumers. Auth remains optional — the public intelligence corpus stays open; only personalization and rate-tiered API access require a session.

stateDiagram-v2
    [*] --> Anonymous

    Anonymous --> Authenticating: User / consumer signs in
    Anonymous --> PublicAccess: Browses open content (no login)
    PublicAccess --> [*]

    Authenticating --> MFAChallenge: Credentials accepted
    Authenticating --> AuthFailed: Bad credentials
    AuthFailed --> Anonymous: Retry (lockout after N attempts)

    MFAChallenge --> Active: MFA verified
    MFAChallenge --> AuthFailed: MFA failed

    Active --> TokenRefresh: Access token near expiry
    TokenRefresh --> Active: Refresh token valid
    TokenRefresh --> Expired: Refresh token invalid

    Active --> Idle: Inactivity timeout
    Idle --> Active: Activity resumes
    Idle --> Expired: Idle limit exceeded

    Active --> SignedOut: User logs out
    Expired --> Anonymous: Re-authenticate
    SignedOut --> Anonymous: Session destroyed

    Active --> Revoked: Security event / admin action
    Revoked --> Anonymous: Tokens invalidated

    note right of Active
        Scopes drive API tier + personalization.
        Minimal PII · GDPR data-minimisation.
    end note

    note right of Revoked
        Compromise → immediate token revocation
        feeds Security Incident Response
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Active,PublicAccess ok
    class Anonymous,Authenticating,MFAChallenge,TokenRefresh,Idle,SignedOut warn
    class AuthFailed,Expired,Revoked err
Copy

15. 🔁 Horizon 3: API Request Lifecycle (API Gateway → Lambda)

The public political-intelligence API (API Gateway → Lambda → DynamoDB / Aurora / Neptune / OpenSearch) serves both anonymous and Cognito-authenticated consumers with tiered rate limits.

stateDiagram-v2
    [*] --> Received

    Received --> AuthZ: Request hits API Gateway
    AuthZ --> RateCheck: Cognito scope / API key resolved
    AuthZ --> Unauthorized: Token invalid (401)

    RateCheck --> Validating: Within tier quota
    RateCheck --> Throttled: Quota exceeded (429)

    Validating --> Dispatch: Schema valid
    Validating --> BadRequest: Schema invalid (400)

    Dispatch --> CacheHit: Lambda checks cache
    Dispatch --> DataLayer: Cache miss

    CacheHit --> Responding: Serve cached payload
    DataLayer --> Responding: Query DynamoDB / Aurora / Neptune / OpenSearch
    DataLayer --> UpstreamError: Backend failure (5xx)

    Responding --> Logged: 200 + provenance headers
    Logged --> [*]

    Unauthorized --> Logged
    Throttled --> RetryAfter: Retry-After header set
    RetryAfter --> Logged
    BadRequest --> Logged
    UpstreamError --> Fallback: Serve last-good cache (flagged)
    Fallback --> Logged

    note right of RateCheck
        Tiers: anonymous · registered · partner
        Token-bucket quotas via Cognito scope
    end note

    note right of Logged
        Every response carries data provenance
        + vintage; full audit trail retained
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Responding,CacheHit,Logged ok
    class Received,AuthZ,RateCheck,Validating,Dispatch,DataLayer warn
    class Unauthorized,Throttled,BadRequest,UpstreamError,Fallback,RetryAfter err
Copy

🔁 Cross-Horizon Lifecycles

16. 🤖 AI/LLM Model Evolution Lifecycle (2026–2037)

Assumptions (per Hack23 AI Policy): major AI model upgrades land roughly annually; competitors (OpenAI, Google, Meta, EU sovereign AI) are evaluated at each release; the architecture is designed to absorb paradigm shifts (quantum AI, neuromorphic computing); all model changes pass through shadow → canary → governance gates with human oversight.

The state machine below treats each model generation as a state the platform occupies, with minor-update, major-upgrade, competitor-switch and AGI-transition transitions.

stateDiagram-v2
    [*] --> CurrentModel

    CurrentModel --> MinorUpdate: Periodic minor release
    MinorUpdate --> Evaluation: Opus 4.x point release
    Evaluation --> Integration: Measurable gain
    Evaluation --> CurrentModel: No significant gain

    Integration --> ShadowTesting: Deploy shadow mode
    ShadowTesting --> CanaryRollout: Quality verified
    ShadowTesting --> Rollback: Quality degraded
    CanaryRollout --> CurrentModel: New model active
    Rollback --> CurrentModel: Revert to previous

    CurrentModel --> MajorUpgrade: Annual major version
    MajorUpgrade --> ArchitectureReview: Opus 5.x / 6.x / 7.x …
    ArchitectureReview --> CapabilityAssessment: Backward compatible
    ArchitectureReview --> PlatformRedesign: Breaking changes
    CapabilityAssessment --> FeatureExpansion: New capabilities unlocked
    FeatureExpansion --> Integration: Enable behind flags
    PlatformRedesign --> MigrationPlanning: Plan substrate update
    MigrationPlanning --> Integration: Migration complete

    CurrentModel --> CompetitorEvaluation: Quarterly review
    CompetitorEvaluation --> ModelSwitch: Competitor superior
    CompetitorEvaluation --> CurrentModel: Current model best
    ModelSwitch --> Integration: Multi-model via Bedrock

    CurrentModel --> AGITransition: 2034–2037 window
    AGITransition --> EnhancedMode: Pre-AGI improvements
    AGITransition --> AutonomousMode: AGI capabilities confirmed
    EnhancedMode --> CurrentModel: Incremental enhancement
    AutonomousMode --> GlobalPlatform: Multi-parliament at scale
    GlobalPlatform --> [*]

    note right of MinorUpdate
        2026: Opus 4.6–4.9 — AI code review,
        test gen, agentic CI/CD (🟢)
    end note

    note right of MajorUpgrade
        2027 Opus 5.x 🔵 · 2028 Opus 6.x 🟣
        2029 Opus 7.x 🟠 · 2030 Opus 8.x 🔴
    end note

    note right of AGITransition
        2031–33 Opus 9–10.x / Pre-AGI ⚪
        2034–37 AGI / Post-AGI ⭐
        Human oversight retained throughout
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class CurrentModel,Integration,CapabilityAssessment,FeatureExpansion,GlobalPlatform ok
    class MinorUpdate,Evaluation,ShadowTesting,CanaryRollout,MajorUpgrade,ArchitectureReview,CompetitorEvaluation,AGITransition,EnhancedMode warn
    class Rollback,PlatformRedesign,MigrationPlanning,ModelSwitch,AutonomousMode err
Copy

16.1 AI Generation → Capability-State Translation (DevSecOps + Product/Data/State lens)

The Hack23 AI Model Evolution roadmap, read as the states each capability domain can reach as the model curve advances:

Each capability advance is gated by the same shadow → canary → governance discipline in §9 and §16; no AI generation auto-promotes to a public state without passing those guards.

17. 🌐 IMF Cache → Aurora Lifecycle Evolution

Baseline: the already-implemented IMF cache lifecycle (vintage-tagged, SHA-256 pinned, supersedes-chain) is documented in STATEDIAGRAM.md §IMF. The machine below preserves that semantics and adds Aurora Serverless v2 transitions (row-level locks, replica lag) as the runtime migrates in v3.0+.

Authoritative hub: analysis/imf/README.md · analysis/imf/agentic-integration.md · analysis/imf/indicators-inventory.json · analysis/imf/data-dictionary.md · .github/aw/ECONOMIC_DATA_CONTRACT.md

stateDiagram-v2
    [*] --> Empty: imf_cache row absent
    Empty --> Fetching: news-* worker requests dataflow/indicator/country
    Fetching --> Fresh: payload ≤ 6 months old · SHA-256 pinned
    Fetching --> Stale: payload > 6 months old
    Fetching --> RateLimited: HTTP 429
    Fetching --> Failed: HTTP 5xx / timeout
    Fresh --> Used: article cites with full confidence
    Stale --> Annotated: staleness_annotated = true required
    Annotated --> Used: article cites with downgraded confidence
    RateLimited --> Backoff: exponential 2^n seconds
    Backoff --> Fetching: retry
    Failed --> CacheFallback: serve last known vintage
    CacheFallback --> Used: article cites with cache-fallback annotation
    Used --> [*]

    state Used {
        [*] --> ProvenanceLogged: article_economic_provenance row inserted
        ProvenanceLogged --> ReplicaSynced: Aurora SvL2 replica catches up (v3.0+)
        ReplicaSynced --> [*]
    }

    note right of Used
        v1.x: static JSON cache + provenance file
        v3.0+: same semantics on Aurora SvL2 —
        row-level locks, replica-lag awareness
    end note

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    classDef err fill:#f44336,stroke:#b71c1c,color:#000000
    class Fresh,Used ok
    class Stale,Annotated,Backoff,CacheFallback warn
    class RateLimited,Failed err
Copy

17.1 WEO vintage transition (April → October cycle)

stateDiagram-v2
    [*] --> WEO_2026_04: April WEO published
    WEO_2026_04 --> WEO_2026_10: October WEO supersedes
    WEO_2026_04 --> Archived: superseded · preserved for audit-trail
    WEO_2026_10 --> WEO_2027_04: April 2027 WEO publishes
    WEO_2026_10 --> Archived
    Archived --> [*]: never deleted (provenance integrity)

    classDef ok fill:#4caf50,stroke:#2e7d32,color:#000000
    classDef warn fill:#ff9800,stroke:#e65100,color:#000000
    class WEO_2026_04,WEO_2026_10,WEO_2027_04 ok
    class Archived warn
Copy

Canonical rule. Every economic claim in a Riksdagsmonitor article cites an IMF dataflow first; World Bank citations are reserved for governance, environment and social residue (the classes IMF does not publish). SCB is the Swedish-specific ground-truth layer. See ECONOMIC_DATA_CONTRACT.md v2.1 for the banned-phrase list and vintage discipline (> 6 mo → annotation).

18. 🛰️ Political-Intelligence Capability State Machines (to 2037)

These lifecycles operationalize the Political-Intelligence Capability Catalog (C1–C32). They add the continuous, runtime state machines the current build-time pipeline lacks: indications & warning, forecast calibration, influence early-warning and cross-source entity resolution. Each terminates in a human-on-the-loop gate; none can reach a published state without an evidence anchor and a party-symmetry check.

18.1 Indications & Warning tripwire lifecycle (C14)

stateDiagram-v2
    [*] --> Monitoring
    Monitoring --> Watch: indicator approaches threshold
    Watch --> Monitoring: relaxes
    Watch --> Elevated: threshold breached
    Elevated --> High: multiple indicators breach
    Elevated --> Triage: human review
    High --> Triage: human review
    Triage --> FalseAlarm: not significant
    Triage --> Confirmed: warning valid
    FalseAlarm --> Tuning: adjust threshold
    Tuning --> Monitoring
    Confirmed --> Retasking: re-task fusion plus forecast
    Retasking --> Briefed: warning note issued
    Briefed --> Monitoring: event window closes
    Briefed --> [*]

    note right of Triage
        Human-on-the-loop mandatory.
        WEP probability attached (ICD-203).
    end note
Copy

18.2 Forecast calibration lifecycle (C13 / C29)

stateDiagram-v2
    [*] --> Open
    Open --> AssumptionsChecked: KAC pass
    Open --> Rejected: KAC fail
    AssumptionsChecked --> Forecast: ensemble produces p plus WEP
    Forecast --> Published: assumptions disclosed
    Published --> AwaitingResolution
    AwaitingResolution --> Scored: event resolves
    Scored --> Calibrated: Brier within band
    Scored --> Drifting: Brier degrades
    Drifting --> Retraining: re-weight ensemble
    Retraining --> Open
    Calibrated --> Open: roll question forward
    Rejected --> [*]

    note right of Scored
        Calibration ledger updated.
        Rolling Brier is a release metric.
    end note
Copy

18.3 FIMI / influence early-warning lifecycle (C20) — aggregate, non-accusatory

stateDiagram-v2
    [*] --> Listening
    Listening --> FrameDetected: narrative frame emerges
    FrameDetected --> Listening: isolated, no coordination
    FrameDetected --> CoordinationSuspected: network-level signal
    CoordinationSuspected --> EthicsGate: privacy plus profiling check
    EthicsGate --> Halted: profiling risk
    EthicsGate --> AttributionAssessed: aggregate-only confirmed
    AttributionAssessed --> Advisory: low/med confidence, WEP
    Advisory --> HumanFraming: analyst contextualizes
    HumanFraming --> Briefed: contextual, not accusatory
    Briefed --> Listening
    Halted --> Listening
    Briefed --> [*]

    note right of EthicsGate
        No citizen profiling.
        Aggregate public discourse only.
    end note
Copy

18.4 Cross-source entity-resolution lifecycle (C1)

stateDiagram-v2
    [*] --> Ingested
    Ingested --> Normalized: provenance stamped
    Normalized --> Matching: deterministic plus embedding
    Matching --> Ambiguous: confidence below cut
    Matching --> Resolved: confidence above cut
    Ambiguous --> HumanReview
    HumanReview --> Resolved: adjudicated
    HumanReview --> Distinct: kept separate
    Resolved --> Linked: canonical ID assigned
    Linked --> Monitoring: watch for new links
    Monitoring --> Matching: new source arrives
    Distinct --> [*]

    note right of HumanReview
        Public records only.
        No private-life inference.
    end note
Copy

18.5 Estimative-product publication lifecycle (C22)

stateDiagram-v2
    [*] --> Drafting
    Drafting --> SATReview: ACH plus devil's advocate
    SATReview --> NeutralityAudit: key judgments drafted
    NeutralityAudit --> Rework: party-asymmetry found
    Rework --> SATReview
    NeutralityAudit --> ICDCheck: symmetric
    ICDCheck --> Rework: uncertainty or sources weak
    ICDCheck --> SignOff: ICD-203 compliant
    SignOff --> Published: human analyst signs
    Published --> Tracked: judgments logged for scoring
    Tracked --> [*]

    note right of ICDCheck
        Sources characterized,
        confidence expressed,
        assumptions distinguished.
    end note
Copy

📋 Future State Summary

🛡️ ISMS Control Mapping

These future state machines remain governed by the Hack23 ISMS. Key control linkage:

Mapped against Secure Development Policy, AI Policy, Threat Modeling and the Classification Framework. See FUTURE_SECURITY_ARCHITECTURE.md and FUTURE_THREAT_MODEL.md for control depth.

📚 Related Documents

Hack23 ISMS Policies

• 🛡️ Secure Development Policy — Architecture documentation requirements
• 🤖 AI Policy — AI usage & human-in-the-loop governance
• 🎯 Threat Modeling Policy — STRIDE / MITRE ATT&CK
• 🏷️ Classification Framework — CIA triad classification

📋 Document Control:
✅ Approved by: James Pether Sörling, CEO
📤 Distribution: Public
🏷️ Classification:
📄 Version: 3.0
📅 Effective Date: 2026-05-31
⏰ Next Review: 2026-08-31
🎯 Framework Compliance:

🔗 Hack23 Ecosystem

🗳️ Empower citizens · 🔍 Strengthen democratic accountability · 🕵️ Illuminate the political process

_{© 2008–2026 Hack23 AB (Org.nr 559534-7807) · Maintainer: James Pether Sörling, CISSP CISM}